Frequently asked questions
What does the composite score mean?
Each signal is in one of four states: ok, watch, warning, or critical. The score is a weighted sum of active states — watch counts 0.5, warning 1.0, critical 1.5, each multiplied by the signal's own weight. It maps to four risk bands calibrated on 1967–2026 history: low (0–5), elevated (5–9), high (9–13), severe (13+). The band, not the raw number, is the estimate.
How is this different from the NY Fed recession probability?
The NY Fed model is a probit on a single input — the 10-year/3-month Treasury spread. Elegant, but one-dimensional: it missed or lagged in cycles where the curve stayed honest while credit and labor cracked first. We include that same spread-probit as one signal among eighteen, so curve risk is priced in without the whole estimate riding on it.
How often is the data updated?
The monitor runs daily and also on the release days of weekly and monthly series (claims on Thursdays, NFCI on Wednesdays, payrolls on the first Friday). Signals only change state when their underlying series publishes — a weekly signal cannot fire mid-week on stale data.
Does a high or severe band mean a recession is certain?
No. These are probabilistic bands, not certainties. Historically, even the severe band has seen recessions fail to materialize within 12 months — and elevated readings have preceded both recessions and soft landings. The bands report what happened after similar configurations in the past, nothing more.
Is this investment advice?
No. Everything here is informational — public indicators run through public rules. Recession timing is one input among many for financial decisions; this site deliberately makes no portfolio claims.
Where does the data come from?
All indicators are public FRED series (Federal Reserve Bank of St. Louis) — each signal page links its source series. Recession dating follows the official NBER chronology. Point-in-time analysis uses ALFRED vintages, the unrevised data as originally released.
What are data revisions and vintages?
Economic series get revised — payrolls, GDP, even unemployment claims are restated months later. A backtest that trains on today's revised history is subtly cheating. We store ALFRED vintages so historical analysis can use the data exactly as it was published on a given date.
Is the service free?
The public channels, daily digests, this site, and the methodology are free. The Telegram bot's paid tier adds instant alerts (instead of the daily digest), personal score thresholds, per-signal filtering, quiet hours, and on-demand commands like /episodes and /analytics.